Further testing confirms that any change of state of is-on sets a reset condition, where both outputs are false and allows in to toggle the output one time only.
So in a reset condition -
If is-on is true, off goes true on any change of state of in.
If is-on is false, on goes true only on a leading edge of in going true and off goes true only on a falling edge of in going false.
So with this in mind I'll now explore the run/step anomaly.